GPU aware MPI implementation OpenMPI vs. MVAPICH2

Accelerated Computing HPC Compilers nvc, nvc++ and nvfortran
1

Hi,

could you please list some of the major issues to consider when making a decision to use either OpenMPI or MVAPICH2. I am not asking which one is better, but rather the best match for my use case.
As an example, an advantage of OpenMPI is probably that it ships with the Nvidia HPC compilers and so there is no need to build it or find a compatible binary version for my platform as with MVAPICH2.
Anything I need to be aware of? We are currently supporting MPICH and INTEL MPI, but as we keep to the standard it probably shouldn’t influence the choice for a GPU aware implmenetation.
I suppose performance is comparable and probably use-case dependent.

Thanks,
LS

2

I suppose performance is comparable and probably use-case dependent.

Probably, but I haven’t done the analysis so don’t really know. Personally I use OpenMPI, but that only because it’s what we ship, but would assume MVAPICH2 works just as well.